Lexical-semantic variables (such as word frequency, imageability and age of acquisition) have been studied extensively in neuropsychology to address the structure of the word production system. The evidence available on this issue is still rather controversial, mainly because of the very complex interrelations between lexical-semantic variables. Moreover, it is not clear whether the results obtained in Indo-European languages also hold in languages with a completely different structure and script, such as Chinese. The objective of the present study is to investigate this specific issue by studying the effect of word frequency, imageability, age of acquisition, visual complexity of the stimuli, grammatical class and morphological structure in word and picture naming in Chinese. The effect of these variables on naming and reading accuracy of healthy and brain-damaged individuals is evaluated using mixed-effect models, a statistical technique that allows to model both fixed and random effects; this feature substantially enhances the statistical power of the technique, so that several variables - and their complex interrelations - can be handled effectively in a unique analysis. We found that grammatical class interacts consistently across tasks with morphological structure: all participants, both healthy and brain-damaged, found simple nouns significantly easier to read and name than complex nouns, whereas simple and complex verbs were of comparable difficulty. We also found that imageability was a strong predictor in picture naming, but not in word naming, whereas the contrary held true for age of acquisition. These results are taken to indicate the existence of a morphological level of processing in the Chinese word production system, and that reading aloud may occur along a non-semantic route (either lexical or sub-lexical) in this language.